Do District Attorneys Represent Their Voters? Evidencefrom California’s Era of Criminal Justice Reform∗

Michael W. Sances†

January 8, 2021

Abstract

Reformers claim district attorneys (DA’s) contribute to mass incarceration by respond-ing to “tough on crime” voters, yet there are no measures of the correspondence be-tween DA and voter preferences. I use a series of criminal justice ballot propositionsin California to explore the link between DA’s and their constituencies. While voterpreferences vary greatly across issues and geography, DA’s almost always take theconservative position. They are on the same side as their voters only half the time,and more liberal electorates are only weakly associated with more liberal DA’s. DA’sare also substantially less representative than other elected officials taking positionson the same issues and facing similar electorates. These results suggest limits to rep-resentation among elected criminal justice officials, and among local elected officialsin general.

∗Draft prepared for the PIPE Collaborative Local Political Economy Symposium on January 15, 2021. Please donot circulate without permission. For helpful comments, I thank Justin de Benedictis-Kessner, Michele Margolis, LisaMiller, Steve Rogers, Chris Warshaw, and seminar participants at the 2020 #CJAPSA Justice and Injustice conference,Temple University’s Public Policy Lab workshop, and the University of Pennsylvania’s American Politics SpeakerSeries.†Assistant Professor of Political Science, Temple University. E-mail: msances@temple.edu. 464 Gladfelter Hall,

Philadelphia, PA, 19122.

Locally elected district attorneys hold significant influence over criminal justice outcomes in

the United States, deciding when to file criminal charges and how severe these charges will be.

A common assumption about these officials is that, as local politicians, they are hypersensitive to

local public opinion. Indeed, such sensitivity was a key goal of state constitutional designers, who

opted for elections from small constituencies rather than appointments to larger constituencies.

Those chosen via the latter, argued one delegate to the 1847 Illinois convention, “cannot have the

necessary acquaintance with the people, their morals, the state of society, and the character of the

parties concerned in the case” (Ellis 2012, 1561).

In the 20th century, the rise of the “tough on crime” prosecutor coincided with a general in-

crease in the public’s punitiveness (Enns 2014). Anecdotal evidence showed incumbent prose-

cutors frequently touting their high conviction rates – or being criticized by challengers for their

insufficiently high rates – while national polls showed large majorities of the public supporting

harsher sentences (Gordon and Huber 2002). More recently, some urban jurisdictions have elected

a new breed of district attorney, one openly committed to using the power of the office to reduce

the prison population (Bazelon and Krinsky 2018). Supporters of this reform movement share the

view that, traditionally, the safest way to win election as prosecutor was to be “tough on crime.”

But as public attitudes on law and order move in a liberal direction, so too will the views of offi-

cials. In the words of journalist Emily Bazelon, the reform movement “definitely starts with local

voters. When local voters rise up and decide that they don’t want a ‘tough on crime’ DA,” policy

will change (Illing 2019).

Do district attorneys (DA’s) actually reflect the criminal justice views of their constituents?

Although polls show the public is punitive on average (Brenan 2020), this does not necessarily

mean that more punitive voters are represented by more punitive district attorneys. One reason it is

difficult to test for such a correspondence directly is that, unlike many other areas of government,

there are no existing measures of the criminal justice policy preferences of district attorneys or their

voters. It is difficult to infer the district attorney’s views from case outcomes, which may result

in part from their anticipation of judge behavior (itself a possible result of constituency opinion)

1

(LaCasse and Payne 1999). Moreover, public views on criminal justice have been overwhelmingly

punitive for many years (Enns 2014). To know if district attorneys become more liberal as voter

opinions change, voter opinions must be observed to vary.

I assess the representation of locally elected district attorneys using the case of California.

Spurred by a 2011 federal Supreme Court decision regarding overcrowded prisons, and in keeping

with a national trend, California considered several criminal justice policy changes between 2012

and 2016, including a reform of its “three strikes” sentencing policy, repeal of the death penalty,

and legalizing adult use of marijuana, to name a few. Many reforms were put before voters as

ballot propositions, allowing me to use election returns to measure voter support for each policy

proposal, and many district attorneys also took positions on these measures, allowing me to assess

representation. And an advantage to examining California in this era is that, unlike during past

eras of reform, there is significant variation in voter preferences, both across issues and across

jurisdictions.1

Contrary to conventional wisdom, I find that district attorneys are systemetically out of step

with their constituents. Despite the heterogeneity in voter preferences, DA’s almost always take

the conservative position on criminal justice issues. The handful of exceptions typically occur

in jurisdictions with very liberal constituencies, though even there DA’s frequently defy public

opinion. On average, DA’s take a position in line with a majority of their voters a little over half

of the time. Moreoever, more liberal areas do not necessarily elect more liberal DA’s, suggesting

that even large shifts in voter opinion are unable to move DA’s from their default, conservative

position. Taking advantage of the fact that other California elected officials – including members

of Congress, state legislators, and city and county officials – also take positions on these same

issues, I also find these other officials are substantially more representative of their constituencies

than DA’s are.1When voters enacted the “three strikes” sentencing regime via Proposition 184 in 1994, every county but San

Francisco voted in favor (see https://elections.cdn.sos.ca.gov/sov/1994-general/ssov/measures-statewide.pdf). Notethat in this paper, I use only cross-sectional variation in voter preferences, given the difficulty of measuring districtattorney positions at just a single point in time. However, one advantage of these data is that voters and officials maybe placed on the same scale, allowing me to gauge congruence as well as responsiveness. For those unfamiliar withthese terms, I discuss them below.

2

These results suggest a common assumption about district attorneys – that they are responsive,

perhaps even overly responsive, to the views of their local electorates – is incorrect. Given the lack

of representation observed here, it may be that the role of public opinion in driving the rise of the

carceral state has been overstated – and that hopes of installing progressive prosecutors, responsive

to newly liberal publics, are misguided. The results also suggest limits to local representation

(Sances 2019), and that models of representation developed at the national level may not fit so

neatly to state and local politics (Rogers 2020).

Studies of Representation

While representation is theoretically multifaceted (Pitkin 1972), almost all empirical studies of

representation compare representative and constituent opinion. While most early studies of rep-

resentation focused on Congress (Miller and Stokes 1963; Achen 1978), later work focuses on

the state (Erikson, Wright, and McKiver 1993; Shor and McCarty 2011; Lax and Phillips 2012)

and, more recently, the city and county levels (Tausanovitch and Warshaw 2014; Warshaw 2019).

Regardless of the level of government, these studies employ measures outlined by Achen (1978).

First, congruence is the degree to which a representative’s position approximates that of their con-

stituents’ (average) opinion. For example, if the average voter in a district prefers a minimum wage

of $15, a representative advocating for a $12 minimum wage is more congruent than a representa-

tive advocating for a $10 minimum wage, as the former representative’s opinion is “closer” to the

district’s voters.

Second, responsiveness is the degree to which a representative’s behavior changes as voter pref-

erences change. For example, if the average voter’s preferred minimum wage changes from $15

to $20, a representative who begins advocating for a higher minimum wage is responsive, while a

representative who does not change their position (or who starts advocating for a lower wage) is

not responsive. Although at a conceptual level, responsiveness captures how the same representa-

tive would change their behavior in response to the same voters changing their preferences, in most

3

empirical applications, analysts rely on cross-sectional comparisons, asking if districts with more

or less liberal voters have more or less liberal representatives. I adopt the same cross-sectional

approach in my paper.2

Because voters are only rarely presented with the same policy choices as their representatives,

we know much more about responsiveness than we do about congruence. Still, most studies of

Congressional representation find a robust connection between representative and aggregate con-

stituency opinion (Bartels 2018). At this level, finding such a relationship is perhaps unsurprising,

given strong electoral incentives (Mayhew 1978) to converge to the median voter in one’s con-

stituency (Downs 1957).

At the state and local level, scholars typically assume similar electoral incentives operate (e.g.,

Tausanovitch and Warshaw 2014). While this may be true in some cases, Rogers (2020) warns that

theories of representation developed at the national level may fail when ported to lower levels. For

instance, Rogers finds representation is much weaker in state legislatures as compared to Congress;

Sances (2020) finds that local government responsiveness is limited to only a few issue areas; and

Thompson (2020) finds elected sheriffs behave similarly in terms of immigration enforcement re-

gardless of their voters’ preferences. In all three of these cases, the authors note key differences

between the institutional environment studied and Congress: a lack of information (Rogers), con-

straints due to state preemption (Sances), and homogeneity in candidate ideology resulting from

professional norms (Thompson).

2Note that there is no a priori reason to expect congruence and responsiveness to go together – in fact, in some casesthey can be at odds. Again take our example of the constituency that prefers a $15 minimum wage, and suppose nowthe representative prefers a $14 minimum wage. Now suppose the constituency becomes more conservative, favoringa $13 minimum wage, and that the representative reacts by changing their position to $4. This representative is quiteresponsive – so much so, in fact, that they have drastically increased the distance between themselves and their voters.Typically, researchers know which constituencies are more or less liberal, and which representatives are more or lessliberal, but they are not able to place voters and representatives on the same scale. When researchers are able to putvoters and officials on the same scale, they sometimes find large gaps (e.g., Bafumi and Herron 2010; Lax and Phillips2012)

4

District Attorney Representation?

Like members of Congress, state legislators, and city mayors, district attorneys are selected via

elections, which in principle allow voters to choose like-minded representatives and induce them

to respond via the threat of removal. Of course, there are several institutional differences that

might prevent elections from leading to district attorney representation. In many states, including

California, these offices are non-partisan, depriving voters of a key signal of candidate preferences.

Also in many states, elections are held separately from races for statewide office; in California they

are mostly held in June of midterm election years. And unlike some other offices, voters might

value technical skill and general performance over ideological views. That said, most of these

objections also hold for city and county elections, where scholars have nonetheless, in general,

found robust evidence of ideological representation (Warshaw 2019).

An additional reason to expect district attorney representation is that a good deal of past work

shows sensitivity to public opinion in a very similar office: local trial judges. Although scholars

rarely measure public opinion directly, a common finding is that more exposure to the public –

whether via election timing (Huber and Gordon 2004; Dippel and Poyker 2019), selection mecha-

nisms (Gordon and Huber 2007; Lim 2013), or press coverage (Lim, Snyder, and Stromberg 2015)

– leads to harsher sentencing behavior. If the median voter is pro-punishment, these results suggest

judges seek to represent voter opinion when they believe voters are watching.3

A handful of studies have similarly tested if district attorneys are more punitive in election

years (Dyke 2007; Bandyopadhyay and McCannon 2014; Nadell, Scaggs, and Bales 2017). While

these papers generally find support for this hypothesis, none of them is able to directly compare

measures of voter opinion to district attorney behavior. The one study that does is Nelson’s (2014)

examination of district attorney behavior in Colorado following the legalization of marijuana. Sim-

ilar to my own paper, Nelson uses the results of the marijuana ballot measure to calculate a measure

3Relaxing the assumption of uniform voter preferences, Park (2017) and Boston and Silveira (2019) both find trialjudges with more liberal (conservative) voters become more liberal (conservative) when more electorally exposed.Finally, several other studies find that when voter preferences become clearer, perhaps as a result of a referendum orrecall campaign, sentencing behavior changes accordingly (Kuklinski and Stanga 1979; Nelson 2014; Gordon andYntiso 2020).

5

of constituency opinion; he finds that after the referendum results were known, DA’s (and judges)

in areas with more supporters alter their sentencing behavior accordingly. However, these results

do not tell us how congruent DA’s are with their voters, and it is unclear whether the findings

extend to other criminal justice issues.

Existing evidence aside, it is worth considering some other key institutional differences be-

tween district attorneys and other offices. Like sheriffs, district attorney candidates almost always

have experience in the criminal justice system, nearly always as prosecutors. This experience

could mean that candidates are uniformly conservative (Thompson 2020). Incumbent DA’s also

have special power over the candidate pool. As shown in Table tk, the surest path to becoming

district attorney is to have served previously as a prosecutor, typically as the chief assistant to the

incumbent. Any deputy making a challenge prior to the incumbent’s chosen retirement date not

only risks losing an election, but their job.

If not the deputies, who else might challenge the incumbent for being out of step with the

voters? Other candidates may have prosecutorial experience from other local or federal offices,

potentially freeing them from the incumbent’s influence – but perhaps also weakening their appeal

to voters skeptical of “outsiders”. Another source of candidates are criminal defense lawyers with

little or no prosecutorial experience. Yet these candidates may be similarly cowed. Would a defense

lawyer who loses to the incumbent district attorney perhaps have a harder time doing their job in

the future? Surely this thought enters any ambitious defense lawyer’s electoral calculus (Wright

2008).4

Data on electoral competition in California races speak to some of the differences between

DA’s and other offices. In Table 1, I show several measures of competition for district attorneys

(top row), CA state assembly members (second row), CA state senators (third row), and US House

members from California (bottom row). To compute these measures, I use election returns from

4Hessick and Morse (2020) argue that a major limitation to electoral competition is the lack of lawyers in smalljurisdictions. In suggestive analysis of national data, they find that the more lawyers living in a jurisdiction, the morelikely it is the incumbent DA sees a challenger. Other possible barriers to DA representation operate at the level of thevoter. Even if a challenger were to inform voters their DA is out of step, voters might opt to keep the unrepresentativeDA if they value the technical skill of prosecuting cases – an experience nearly exclusive to the DA and their deputies.

6

Office Total Incumbent Incumbent Incumbent Incumbentraces ran opposed won if ran won if opposedN % % % %

(N) (N) (N) (N)

District attorney 116 83 41 88 69(96) (39) (84) (27)

State assembly 160 81 94 95 94(129) (121) (122) (114)

State senate 40 57 96 91 91(23) (22) (21) (20)

US House 106 92 100 95 95(98) (98) (93) (93)

Table 1: Comparing electoral competition across offices, 2014-2018.

both the 2014 and 2018 races.

Table 1 shows that incumbents run at about the same rate as these other offices (first column),

though they are opposed far less (second column).5 Just 41% of district attorneys faced any op-

ponent in an election cycle, compared to 94% of state assembly members, 96% of state senators,

and 100% of US House members.6 Conditional on running, though, incumbent district attorneys

actually win less often than legislators (88% of the time), and win even less (69% of the time)

when they face a challenger. Thus, a distinguishing feature of district attorney elections is the lack

of opposition. With no challenger to seriously threaten their job, district attorneys may have little

need to respond to their voters.

Criminal Justice Reform in California

Starting in 2011, criminal justice policy in California has generally moved in the direction of lib-

eral reform (Arango 2019). A catalyst was a 2011 federal Supreme Court ruling, which ordered

5California has term limits for state legislators, which likely explains the relatively low share of incumbent statesenators who run for re-election.

6I count both primary and general election challengers for these figures.

7

the state to drastically reduce its overcrowded prison population (Liptak 2011). In response, state

lawmakers enacted “realignment”, shifting thousands of state inmates convicted of “nonviolent and

nonserious” offenses to counties (VanSickle and Villa 2018). In addition to legislative action, re-

form has also taken place via direct democracy. For example, in 2014, voters approved Proposition

47, which re-classified many felony crimes to misdemeanors with more lenient penalties; and in

2016, voters approved Proposition 57, which gave thousands of state prisoners a chance at parole

(VanSickle and Villa 2018). Not all these attempts at reform succeeded: for instance, in both 2012

and 2016, voters turned down efforts to abolish the death penalty. But overall, the shift in policy

was so great that by 2018, a Los Angeles Times headline claimed that the state had “transformed

its justice system” since 2011 (VanSickle and Villa 2018).

To measure constituency opinion on criminal justice policy, I examine relevant ballot propo-

sitions considered in the 2012, 2014, and 2016 November elections.7 One reason for selecting

this time span is that it is after the 2011 Supreme Court decision and the start of the state’s liberal

shift on justice policy. Another reason is that it is difficult to determine the positions of district

attorneys, and it is more difficult the further back in time we go. By my count, Californians voted

on a total of nine propositions related to criminal justice policy over this period, and I summarize

them below.8

• Death penalty repeal (Proposition 34, 2012): Would have repealed the death penalty and

replaced it with life without parole, and would have applied retroactively to those currently

on death row. Failed with 48% in favor (Ballotpedia Authors 2020a).

• Human trafficking increase sentences (Proposition 35, 2012): Increased penalties for

human trafficking convictions, required traffickers to register as sex offenders, and required

sex offenders to disclose Internet activity. Passed with 81% in favor (Ballotpedia Authors

2020b).7Senate Bill 202, enacted in 2011, mandates that ballot propositions must be held in November of statewide election

years (i.e., even years) (Ballotpedia 2020j). There were no criminal justice-related propositions on the ballot in 2018.8The bolded phrase is my own shorthand for each proposition, not the official proposition title.

8

• Soften three-strikes (Proposition 36, 2012): The existing three-strikes law resulted in a

life sentence on the third offense; this required the third offense to be “serious or violent.”

Passed with 69% in favor (Ballotpedia Authors 2020c).

• Nonviolent offenses to misdemeanors (Proposition 47, 2014): Mandated misdemeanor

sentences instead of felonies for some drug and property crimes. Passed with 60% in favor

(Ballotpedia Authors 2020d).

• Increase parole for nonviolent felons (Proposition 57, 2016): Allowed those convicted

of “nonviolent felonies” to be considered for parole once prison term for main offense is

complete. Passed with 65% in favor (Ballotpedia Authors 2020e).

• Death penalty repeal (Proposition 62, 2016): Essentially the same as Proposition 34 in

2012. Failed with 47% in favor (Ballotpedia Authors 2020f).

• Gun control (Proposition 63, 2016): Instituted a variety of gun control measures, including

requiring background checks and state authorization for buy ammunition, banning high-

capacity magazines, and requiring most ammunition sales sales to be made through licensed

vendors. Passed with 63% in favor (Ballotpedia Authors 2020g).

• Legalize recreational marijuana (Proposition 64, 2016): Legalized all marijuana use un-

der state law. Passed with 57% in favor (Ballotpedia Authors 2020h).

• Death penalty speed process (Proposition 66, 2016): Made a number of changes to pro-

cedures for death penalty convictions, including limiting the amount of time for legal chal-

lenges to five years. Passed with 51% in favor (Ballotpedia Authors 2020i).

Measuring Constituency Opinion

I follow several existing studies of representation that use constituency returns on ballot proposi-

tions as an alternative to survey measures of voter opinion (Brunner, Ross, and Washington 2013;

9

Masket and Noel 2012; Matsusaka 2017; Rogers 2017; Giger, Klüver, and Witko 2020). While

most of these past studies go on to match constituency returns with actual legislative votes on the

issue, as I explain below, I instead use endorsements by public officials. Before doing so, I describe

the constituency measures.

All district attorneys in California serve at the county level. As such, I measure constituency

opinion on each proposition using county election returns, obtained from the Secretary of State’s

web site.9 In Figure 1, I map how voters in each county voted on each proposition, with darker

shadings indicating that a majority over voters in that county cast a vote for the liberal position.

(For almost all of the propositions, a “yes” vote corresponds with the liberal position; the excep-

tions are the human trafficking and death penalty speed propositions.) These maps show consid-

erable heterogeneity in voter liberalism across issues and geography. For example, a majority of

voters in all counties voted to increase penalties for human trafficking – a conservative position –

while also voting to decrease penalties for “third strikers” – a liberal position. Other issues, such

as the death penalty, see a small group of coastal counties voting liberally while the rest of the state

votes conservative.

Measuring District Attorney Positions

I depart from existing measures of judicial and district attorney responsiveness by using publicly

expressed policy positions, in the form of pre-election ballot proposition endorsements, rather

than administrative data on case filings or outcomes. While no measure is perfect, I argue public

positions have several advantages over administrative data. First, I can place endorsements on the

same scale as voter opinions, allowing me to assess congruence as well as responsiveness. Second,

I can also place them on the same scale as legislators, allowing me to benchmark DA representation

against legislator representation. Third, case data are the result of a complex strategic interaction

involving the police (Knox, Lowe, and Mummolo 2020), the DA (more likely, an assistant DA and

9See https://www.sos.ca.gov/elections/prior-elections/statewide-election-results/.

10

Figure 1: California counties with a majority voting on liberal side, by proposition. Light graycounties saw less than 50% take a liberal position, while black counties saw more than 50% take aliberal position.

11

Figure 2: California counties represented by a district attorney (DA) taking a liberal position, byproposition. Light gray counties are represented by a DA taking a conservative position, whileblack counties are represented by a DA taking a liberal position. Whited out areas denote countieswhere the DA did not take a position.

12

not the actual elected official), defense attorneys, and judges; in particular, the veto power of judges

(LaCasse and Payne 1999) may prevent the DA from acting in accordance with voter wishes.10

I obtained DA positions on four of the nine propositions from the American Civil Liberties

Union of California. As part of a broader effort to educate voters in district attorney elections, the

ACLU set up a web site, https://meetyourda.org/, publicizing the positions of all CA district attor-

neys on Proposition 36 in 2012 (softening three strikes), Proposition 47 in 2014 (change felonies

to misdemeanors), Proposition 57 in 2016 (increase parole), and Proposition 64 in 2016 (legalize

marijuana). For each proposition, almost all district attorneys are coded as supporting, opposed,

or neutral. I use the data from this web site as a starting point for coding DA positions. To sim-

plify the analysis and to ensure comparability with other offices, I code any neutral positions as

missing.11

For the remaining five propositions, I relied mainly on the endorsements pages of pro- and anti-

campaign web sites, which I found using the Wayback Machine tool at the Internet Archive (see

https://archive.org/web/). These campaign web sites typically list public officials who agree with

the campaign’s position.12 To supplement these lists, I searched the web and newspaper databases

to find instances of specific district attorney’s taking positions.13 Last, I also searched for district

attorney names on lists of donors to pro- and anti-campaign committees for each proposition, via

the California Secretary of State’s web database (see http://cal-access.sos.ca.gov/).

Figure 2 plots district attorney positions by county for each of the propositions. Mirroring the

10The use of administrative data also requires numerous researcher choices about how to code different cases, whatcases to exclude, and what constitutes a more punitive outcome, raising the risk of specification search (Dippel andPoyker 2019). A disadvantage of using endorsements is that officials are under no obligation to offer an opinion, andthe lack of an opinion could be strategic. Moreover, endorsements do not necessarily reflect policy decisions. Thatsaid, roll call votes cast in legislatures are also strategic, and may also not reflect actual policy changes.

11On most issues, only a handful of district attorneys take neutral positions. Exceptions are Proposition 36 in 2012(three strikes), where 24 are neutral, and Proposition 64 in 2016 (legalize marijuana), where 43 are neutral. See theOnline Appendix for results using alternative codings of these neutral positions.

12For two propositions, there was one side where I could not locate an official list of supporters: the anti-campaignfor Proposition 36 in 2012, as the campaign site was not archived on Wayback; and the anti-campaign for Proposition35 in 2012, as the campaign site appeared to be a blog and did not list any endorsements.

13First, I determined the name of the incumbent district attorney for each of the three proposition years (2012,2014, and 2016) for all 58 California counties. Second, I searched for documents including both the district attorney’sname, and the phrase “proposition [#]” or “prop [#]” for each proposition. I performed these searches using Google,newspapers.com, and the ProQuest US Newsstream database.

13

scheme in Figure 1, counties with district attorneys taking liberal positions are shaded dark gray,

while those with conservative positions are shaded light gray; counties that are whited out are

areas where the district attorney either took a neutral position, or I was not able to find any public

position.14

In contrast to the variety of opinion seen for voters, Figure 2 shows remarkable consistency

in district attorney positions, regardless of issue or geographic area. For a typical proposition, all

district attorneys with positions are united in taking a conservative stance, with perhaps one or two

liberal dissenters. The sole exception is seen in the bottom left panel, concerning the gun control

measure. In this case, five district attorneys took liberal positions and four took conservative

positions. The relative consistency of district attorney positions, compared to the great variety

in voter opinion, constitutes preliminary evidence of a lack of representation.

Measuring Representation

I use two standard metrics to quantify the extent of district attorney representation (Achen 1978;

Lax and Phillips 2012). First, for each county I record whether the DA takes a position that is

consistent with, or on the same “side” as, their constituency. I refer to this measure as congruence,

14District attorneys are actually somewhat more likely to take public positions on these propositions as comparedto legislators. While there are a total of 173 district-based legislators in the state (80 State Assembly, 40 State Senate,and 53 US House), for the average proposition, 25 of these legislators take a position. In contrast, for the averageproposition, 29 of 58 district attorneys take a position. To the extent that taking public positions is strategic, makingcomparisons across offices requires the assumption that any bias is independent of the office. One way this mightfail is if DA candidates feel compelled to offer a position on criminal justice propositions, and so more easily revealthemselves as out of step, while other offices do not feel compelled to offer a position, and so only those who arecongruent do so. However, while DA’s are more likely to take a position than other offices, they do not appear to feelcompelled to do so, as missingness is still frequent.

14

and compute it as

Congruencei j =

1 if Votei j > 50 and District Attorneyi j = 1

1 if Votei j < 50 and District Attorneyi j = 0

0 if Votei j < 50 and District Attorneyi j = 1

0 if Votei j > 50 and District Attorneyi j = 0

where Votei j represents the liberal vote share, and District Attorneyi j is an indicator for the DA

taking the liberal position, for each district attorney i and for each proposition j. Second, I ask if,

cross-sectionally, the tendency of a district attorney to adopt a liberal position is associated with

voter liberalism. I refer to this as responsiveness, and measure it using the slope (β j) from the

regression

District Attorneyi j = α j +β j ∗Votei j + εi j

Before computing these statistics numerically, I visualize the relationship between positions

and opinions, plotting one against the other, in Figure 3. Consistent with the prior two figures, for

most propositions we see observations concentrated on the bottom portion of the plots, indicating

conservative district attorney positions, regardless of what side of the horizontal axis voters are

on. Points in the lower left and upper right quadrants of the figure indicate congruence. For

instance, in the top left panel (the first death penalty measure), we see evidence of congruence,

as constituencies to the left of the vertical dashed line are represented by conservative district

attorneys, while the lone liberal district attorney represents a very liberal district. On other issues,

such as marijuana repeal shown in the bottom middle panel, we see virtually no congruence, as all

but one constituency votes liberally while no district attorney takes a liberal stance.

Figure 3 also shows preliminary evidence of responsiveness, but only on some issues. Each

panel includes a solid line representing the regression equation given previously. In most cases,

the slope of the line is positive, though a lack of variation in district attorney positions means the

line can not be computed for some of the issues. Even when the lines are positive, furthermore, the

15

Congruence=82%

Con

Lib

0 25 50 75 100

Prop 34 (2012)Death penalty repeal

94%Con

Lib

0 25 50 75 100

Prop 35 (2012)Human trafficking increase sentences

19%Con

Lib

0 25 50 75 100

Prop 36 (2012)Soften three-strikes

38%Con

Lib

0 25 50 75 100

Prop 47 (2014)Nonviolent offenses to misdemeanors

29%Con

Lib

0 25 50 75 100

Prop 57 (2016)Increase parole for nonviolent felons

84%Con

Lib

0 25 50 75 100

Prop 62 (2016)Death penalty repeal

89%Con

Lib

0 25 50 75 100

Prop 63 (2016)Gun control

13%Con

Lib

0 25 50 75 100

Prop 64 (2016)Legalize recreational marijuana

75%Con

Lib

0 25 50 75 100

Prop 66 (2016)Death penalty speed process

DA

posi

tion

Voter liberalism

Figure 3: District attorney positions and constituency opinion.

16

p=0.065

p=0.000

p=0.000

p=0.000

0

.25

.5

.75

1Pr

opor

tion

cong

ruen

t pos

ition

s

DA's Local officials State legislators US House members

Figure 4: Congruence by office. Bar heights are the share of observations where officials take acongruent position. P-values from the test that the true share is 50% (and thus “random” congru-ence) are displayed above each bar.

relationships are weak, suggesting that even large liberal shifts in constituency opinion would not

shift district attorney positions by very much.

Congruence Across Issues and Offices

Figure 4 summarizes DA congruence, plotting the proportion of the time DA’s take positions that

are congruent with their constituents. Across 259 positions, DA’s are congruent with their voters

about 58% of the time. This is substantively not much higher than 50%, which is the degree we

would see DA’s congruent with their voters if incumbents simply flipped a coin. Statistically, we

can’t reject the null hypothesis that the true share is 50% at conventional levels.15

15To conduct this test, I use the block bootstrap, blocking at the level of the district, with 1,000 replications.

17

To put the figure of 58% in perspective, the remaining bars re-compute the proportion of con-

gruent positions for other offices. Across 212 instances of position-taking, local officials (mayors,

city councils, and county supervisors) are congruent 67% of the time. Across 173 positions, CA

state legislators are congruent 73% of the time. And across 50 positions, US House members from

California are congruent 86% of the time. In contrast to the figures for DA’s, for all three of these

figures we can easily reject the null hypothesis of “random” congruence at conventional levels.16

As another way of exploring the coincidental nature of any DA congruence, Figure 5 plots

an indicator for congruence against voter liberalism, pooling all issues for each type of office.

Points in these figures are jittered for legibility, and the solid lines represent moving averages

from quadratic regressions. Interestingly, for the three non-DA offices, the data have a “U” shape

suggesting a dip in congruence in more divided electorates. This suggests that officials in these

districts seek to adopt positions in line with their voters, but that they find it harder to discern voter

opinion (or easier to shirk) when voters are not overwhelmingly on one side. In contrast, for DA’s

there is no such apparent curvature. Instead, whether a DA is congruent is almost entirely a result

of whether the constituency is conservative. If so, voters get a DA in line with their views. If not,

they do not.

Responsiveness Across Issues and Offices

The evidence in Figure 3 suggests DA’s are weakly responsive on some issues, and entirely unre-

sponsive on others. To summarize these patterns, I pool all issues into a single regression, with

results repoted in Table 2. As with the analysis of congruence, I also perform the analysis for other

offices for the purposes of comparison.

16For the non-DA offices, I rely exclusively on lists of endorsements, foregoing the more extensive searches of theweb and newspaper databases, to measure elite positions. To measure voter positions, for 2014 and 2016 measuresI rely on official election results released at the level of legislative districts. Legislative districts were redrawn forthe 2012 election, meaning the legislative districts given on the 2012 election results were for future, not current,constituencies. Thus to calculate voter opinion in 2012, I construct my own measure of constituency opinion byaggregating precinct-level results to pre-2012 legislative districts using GIS software. I obtained precinct-level resultsand pre-2012 district maps from the Statewide Database at https://statewidedatabase.org/.

18

0

1

0

1

0 25 50 75 100 0 25 50 75 100

DA Local

State legislator US House member

Cong

ruen

t

Voter liberalism (%)

Figure 5: Congruence against voter liberalism, by office. Solid lines represent quadratic fits. Pointsare jittered for legibility. For most offices, congruence is less likely when voter opinion is moredivided, as reflected by the “U” shape. For DA’s, congruence is simply less likely when voters aremore liberal, as reflected by the relative lack of curvature.

Column 1 of Table 2 confirms that DA’s are responsive, in the sense that more liberal electorates

are represented by more liberal DA’s. Specifically, for every one percentage point increase in

voter liberalism, the probability the DA adopts the liberal position increases by one half of one

percentage point. While this estimate is statistically significant at the 5% level, it is substantively

small, suggesting that even a shift from 0% to 100% liberalism would only make a DA only 50

percentage points more likely to adopt the liberal position. Given the “default” DA position is

conservative, this again suggests that liberal voters get something akin to random representation

from their DA’s, in line with their views maybe half the time. In contrast, all of the other offices

19

(1) (2) (3) (4)District attorneys Local officials State legislators US House members

Voter liberalism 0.005∗ 0.016∗∗∗ 0.018∗∗∗ 0.021∗∗∗

(0.002) (0.001) (0.001) (0.003)

Constant -0.208∗ -0.531∗∗∗ -0.581∗∗∗ -0.440∗

(0.093) (0.055) (0.059) (0.194)

Observations 259 212 173 50Clusters 57 164 82 28

Table 2: Responsiveness: Does the official take the liberal position? Cell entries reflect how theprobability of taking the liberal position changes as the proportion of liberal voters increases byone percentage point, with standard errors in parentheses. Standard errors computed using theblock bootstrap, clustering on office, using 1,000 replications. * p<0.05, ** p<0.01, *** p<0.001

are more responsive (and the estimates are far more precise). In particular, local officials and state

legislators are over three times as responsive as DA’s, while US House members are four times as

responsive.17

Conclusion

For decades, the conventional wisdom has been that candidates for district attorney must style

themselves as “tough on crime” to appease voters. In our current era of reform, the DA’s supposed

over-responsiveness to punitive electorates has been blamed for the rise of the carceral state, with

the DA’s continued responsiveness to a new, reform-minded electorate seen as a path to criminal

justice reform. Yet despite the presumption that DA’s seek to represent their electorates, no existing

study has examined this question directly. Using a series of ballot propositions on criminal justice

reform in California, I conduct this comparison for the first time. Contrary to the conventional

17One difference between district attorneys and these other elected offices is election timing. Typically in California,there is a first round of elections for district attorney in June of midterm election years (e.g., 2014, 2018, etc.). If nocandidate receives more than 50% of the vote, there is then a runoff in November. My measures of voter preferencescome from November elections in even-numbered years. So, it could be that district attorneys are responding to theirJune electorates, which are more conservative. (Most CA municipal races are in November of even years, so using thelocal races does not account for this explanation; see Marschall and Lappie 2014.) However, given district attorneysshould know they could potentially face voters in a November runoff, then they would pay attention to the preferencesof November voters if they seek to represent their constituents.

20

wisdom, I find DA’s are in fact systematically unrepresentative of their voters, both in terms of

congruence and in terms of responsiveness. What is more, I find DA’s are abnormally unrepresen-

tative when compared to other elected officials at the local, state, and federal levels. My results

suggest that while DA’s own preferences may have played a role in the rise of the carceral state,

this may have been in spite of, rather than because of, their constituents’ opinions.

Beyond district attorney elections, my results suggest limits to electoral accountability among

local criminal justice officials, echoing recent work by Thompson (2020) on elected sheriffs. Fu-

ture research should further explore whether this pattern holds; if so, it may mean criminal justice

reform wll have a better chance of succeeding at the state and federal level, rather than the local

level. Future work can also explore the barriers to representation among elected criminal justice of-

ficials, and the extent to which they are driven by features of the candidate pool (e.g., professional

norms that encourage conservative policy views) or by voters (e.g., a preference for technical skill

over issue congruence).

These results also speak to the need for scholars of subnational politics to pay special attention

to the distinct institutional environments often faced by the officials we study (Rogers 2012). For

some local offices, these environments may be similar enough to those at the national level that

theories of representation developed for Congress may be an adequate guide. But for other offices,

such theories may fail.

21

References

Achen, Christopher H. 1978. “Measuring Representation.” American Journal of Political Science

22(3): 475-510.

Arango, Tim. 2019. “In California, Criminal Justice Reform Offers a Lesson for the Nation.”

New York Times (Jan. 21). https://www.nytimes.com/2019/01/21/us/california-incarceration-

reduction-penalties.html

Bafumi, Joseph, and Michael C. Herron. "Leapfrog representation and extremism: A study of

American voters and their members in Congress." American Political Science Review (2010):

519-542.

Ballotpedia Authors 2020a. tk

Ballotpedia Authors 2020b. tk

Ballotpedia Authors 2020c. tk

Ballotpedia Authors 2020d. tk

Ballotpedia Authors 2020e. tk

Ballotpedia Authors 2020f. tk

Ballotpedia Authors 2020g. tk

Ballotpedia Authors 2020h. tk

Ballotpedia Authors 2020i. tk

Ballotpedia Authors 2020j. tk

Bandyopadhyay, Siddhartha, and Bryan C. McCannon. "The effect of the election of prosecutors

on criminal trials." Public Choice 161, no. 1-2 (2014): 141-156.

Bartels, Larry M. Unequal democracy: The political economy of the new gilded age. Princeton

University Press, 2018.

Bazelon, Emily, and Miriam Krinsky. 2018. “A Wave of New Prosecutors Means Justice.” New

York Times (December 13). Page A29. https://www.nytimes.com/2018/12/11/opinion/how-

local-prosecutors-can-reform-their-justice-systems.html

Boston, J. O. S. H. U. A., and BERNARDO S. Silveira. The electoral connection in court: How

22

sentencing responds to voter preferences. Technical report, Mimeo: UCLA, 2019.

Brenan, Megan. 2020. “Fewer Americans Call for Tougher Criminal Justice System.” Gallup.com

(November 16). https://news.gallup.com/poll/324164/fewer-americans-call-tougher-criminal-

justice-system.aspx

Brunner, Ross, and Washington 2013. tk

Dippel, Christian, and Michael Poyker. 2019. “How Common are Electoral Cycles in Criminal

Sentencing?” No. w25716. National Bureau of Economic Research.

Downs, Anthony. "An economic theory of democracy." (1957): 260-276.

Dyke, Andrew. "Electoral cycles in the administration of criminal justice." Public Choice 133, no.

3-4 (2007): 417-437.

Ellis, Michael J. 2012. “The origins of the elected prosecutor.” Yale Law Journal 121(6): 1528-

1570.

Enns, Peter K. 2014. “The public’s increasing punitiveness and its influence on mass incarceration

in the United States.” American Journal of Political Science 58(4): 857-872.

Erikson, R.S., Wright, G.C., Wright, G.C. and McIver, J.P., 1993. Statehouse democracy: Public

opinion and policy in the American states. Cambridge University Press.

Giger et al 2020. tk.

Gordon, Sanford C., and Gregory A. Huber. 2002. “Citizen oversight and the electoral incentives

of criminal prosecutors.” American Journal of Political Science 46(2): 334-351.

Gordon, Sanford C., and Gregory A. Huber. 2007. “The effect of electoral competitiveness on

incumbent behavior.” Quarterly Journal of Political Science 2(2): 107-138.

Gordon, Sanford C., and Sidak Yntiso. "A Silent Corrupting Force? Criminal Sentencing and the

Threat of Recall." Criminal Sentencing and the Threat of Recall (June 4, 2020) (2020).

Hessick, C.B. and Morse, M., 2019. Picking Prosecutors. Iowa L. Rev., 105, p.1537.

Huber, Gregory A., and Sanford C. Gordon. "Accountability and coercion: Is justice blind when it

runs for office?." American Journal of Political Science 48, no. 2 (2004): 247-263.

Illing, Sean. 2019. “How bad prosecutors fuel America’s mass incarceration problem.” Vox.com

23

(May 15). https://www.vox.com/future-perfect/2019/5/15/18523984/mass-incarceration-prosecutors-

emily-bazelon-charged

Knox, Lowe, and Mummolo 2020

Kuklinski, James H., and John E. Stanga. "Political participation and government responsiveness:

The behavior of California superior courts." The American Political Science Review (1979):

1090-1099.

LaCasse, Chantale, and A. Abigail Payne. 1999. “Federal sentencing guidelines and mandatory

minimum sentences: Do defendants bargain in the shadow of the judge?” Journal of Law and

Economics 42(S1): 245-270.

Lax, Jeffrey R., and Justin H. Phillips. "The democratic deficit in the states." American Journal of

Political Science 56, no. 1 (2012): 148-166.

Lim, Claire SH. "Preferences and incentives of appointed and elected public officials: Evidence

from state trial court judges." American Economic Review 103, no. 4 (2013): 1360-97.

Lim, Claire SH, James M. Snyder Jr, and David Strömberg. "The judge, the politician, and the

press: newspaper coverage and criminal sentencing across electoral systems." American Eco-

nomic Journal: Applied Economics 7, no. 4 (2015): 103-35.

Masket and Noel 2012. tk.

Matsusaka 2017. tk

Mayhew, David R. 1978. Congress: The electoral connection. Yale university press.

Miller, Warren E., and Donald E. Stokes. 1963. “Constituency influence in Congress.” American

Political Science Review 57(1): 45-56.

Nadel, Melissa R., Samuel JA Scaggs, and William D. Bales. "Politics in punishment: The effect of

the state attorney election cycle on conviction and sentencing outcomes in florida." American

journal of criminal justice 42, no. 4 (2017): 845-862.

Nelson, Michael J. "Responsive justice? Retention elections, prosecutors, and public opinion."

Journal of Law and Courts 2, no. 1 (2014): 117-152.

Park, Kyung H. "The impact of judicial elections in the sentencing of black crime." Journal of

24

Human Resources 52, no. 4 (2017): 998-1031.

Pitkin, Hanna Fenichel. 1972. The Concept of Representation. Berkeley: University of California

Press.

Rogers 2017. tk.

Rogers, Steven. 2020. Accountability in State Legislatures. Book manuscript.

Sances, Michael W. 2019. “When Voters Matter: The Limits of Local Government Responsive-

ness.” Forthcoming at Urban Affairs Review. September 2019.

Shor, Boris, and Nolan McCarty. "The ideological mapping of American legislatures." American

Political Science Review (2011): 530-551.

Tausanovitch, Chris, and Christopher Warshaw. "Representation in municipal government." Amer-

ican Political Science Review (2014): 605-641.

Thompson, Daniel M. "How partisan is local law enforcement? Evidence from sheriff cooperation

with immigration authorities." American Political Science Review 114, no. 1 (2020): 222-236.

Vansickle, Abbie, and Manuel Villa. 2018. “California transformed its justice system. But now

crime is up, and critics want rollbacks.” Los Angeles Times (Dec. 20).

Warshaw, Christopher. "Local elections and representation in the United States." Annual Review

of Political Science (2019).

Wright, R.F., 2008. How prosecutor elections fail us. Ohio St. J. Crim. L., 6, p.581.

25

Appendix

(1) (2) (3) (4)District attorneys Local officials State legislators US House members

Voter liberalism -0.031∗ -0.049∗∗∗ -0.052∗∗∗ -0.027(0.015) (0.005) (0.007) (0.015)

Liberalism squared 0.000 0.000∗∗∗ 0.000∗∗∗ 0.000∗

(0.000) (0.000) (0.000) (0.000)

Constant 1.765∗∗∗ 1.805∗∗∗ 2.017∗∗∗ 1.436∗∗∗

(0.290) (0.080) (0.149) (0.335)

Observations 259 212 173 50Clusters 57 164 82 28

Table 3: Congruence: Does the official take the position of a majority of constituents? Cell entriesreflect how the probability of congruence changes as the proportion of liberal voters increases byone percentage point, with standard errors in parentheses. Standard errors computed using theblock bootstrap, clustering on office, using 1,000 replications.

26